Batch output tensor of the preprocessing for the pose classification pipeline

Please provide complete information as applicable to your setup.

• Hardware Platform (GPU)
• DeepStream Version: 6.3
• Issue Type( questions)

Hello I am working on the pose classification from the skeleton for the custom model of different key points and for the full pipeline.

I am reading the preprocessing prepare the tensor buf for the third model pose classificiton of TAO toolkit.

Link []

The question is about the output tensor batching.

I saw that the preprocessing is working on only batch of the input but the output tensor is only one first unit?

Feeding the batch unit with multi object id

Preparing only tensor data for the first object id only?

Please confirm is that right? and if so how to generete the tensort output in batch with multi objects (object id)?

Thank you so much!

This sample only considered the pose classifier model batch size is 1 situation. We will modify for arbitrary model batch size.

Thank you so much!! I got it

Do you mean Deepstream had a plan for the batching preprocessing tensor for the further release?

Do you have any suggestions to do with current status?
I am going to add link multi-processing from the tee and then use the second streammux to batch it feed into the pose classification. Is that possible to work with the second streammux in the same main pipeline?


It is open source, you can add the function by yourself.
Current sample configuration is batch size 1, it can work. If you want the SGIE to work with arbitrary batch size, you can add the batch loop in the code. The key is to generate the tensor data with the same track id.

1 Like

I don’t understand what is the second streammux for? Why should you add the second streammux?

Thank you! I got it. the response is helpful. I will close the topic!

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.